Research map / Image generation / Text-to-image diffusion
DiffGraph: An Automated Agent-driven Model Merging Framework for In-the-Wild Text-to-Image Generation
Zhuoling Li, Hossein Rahmani, Jiarui Zhang, Yu Xue, Majid Mirmehdi, Jason Kuen, Jiuxiang Gu, Jun Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add node registration (text node features) (ablation) | DiffGraph · DABench (SD15) | ImageReward: 31.04 -> 73.11 | Improved |
| use fixed text features instead of learnable registration embeddings (ablation) | DiffGraph · DABench (SD15) | ImageReward: 33.36 -> 73.11 | Improved |
| add node calibration (quality-score edge features) (ablation) | DiffGraph · DABench (SD15) | ImageReward: 11.92 -> 73.11 | Improved |
| use measured edge features instead of learnable ones (ablation) | DiffGraph · DABench (SD15) | ImageReward: 19.63 -> 73.11 | Improved |
| add Expert Selection Agent (ablation) | DiffGraph · DABench (SD15) | ImageReward: 16.12 -> 73.11 | Improved |
| select experts with ESA instead of random activation (ablation) | DiffGraph · DABench (SD15) | ImageReward: 15.36 -> 73.11 | Improved |
| use VGAE merging planner instead of equal weights (ablation) | DiffGraph · DABench (SD15) | ImageReward: 13.29 -> 73.11 | Improved |
| use VGAE instead of LLM-produced coefficients (ablation) | DiffGraph · DABench (SD15) | ImageReward: 18.00 -> 73.11 | Improved |
| use text/score node features instead of expert parameters (ablation) | DiffGraph · DABench (SD15) | ImageReward: 26.62 -> 73.11 | Improved |
| add new experts via registration+calibration without retraining (ablation) | DiffGraph · DABench (SD15) | ImageReward: 43.96 -> 69.64 | Improved |
| add new experts post hoc (ablation) | ESA*+K-LoRA · DABench (SD15) | ImageReward: 13.91 -> 19.33 | Improved |
| graph-based merging vs best baseline DiffAgent | DiffGraph · DABench (FLUX) | ImageReward: 100.20 -> 136.62 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Direct — DiRecT: Safe Diffusion-Based Planning via Receding-Horizon Denoising (NeurIPS 2026) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EC-DIT: Scaling Diffusion Transformers with Adaptive Expert-Choice Routing (ICLR 2025 · similarity 0.86)
- DiffAgent: Fast and Accurate Text-to-Image API Selection with Large Language Model (CVPR 2024 · similarity 0.86)
- MCCD: Multi-Agent Collaboration-based Compositional Diffusion for Complex Text-to-Image Generation (CVPR 2025 · similarity 0.85)
- Diff-Aid: Inference-time Adaptive Interaction Denoising for Rectified Text-to-Image Generation (NeurIPS 2026 · similarity 0.85)
- Progressive Text-to-Image Diffusion with Soft Latent Direction (AAAI 2024 · similarity 0.85)
- EvolveDirector: Approaching Advanced Text-to-Image Generation with Large Vision-Language Models (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
